JPH1011576A - Component shape recognizing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To speedily and precisely recognize a shape by encoding picture data of a fetched component picture, inputting encoded data to a neural network and outputting a category from the neural network so as to reduce an arithmetic quantity between picture information. SOLUTION: The picture of the component is inputted from a picture input system 7. The inputted picture is temporarily read in a frame memory 4 and read at a proper timing by the command of a CPU 2 to extract the picture of a specified edge or corner part. Data of these edge or corner part pictures is outputted to each small area previously fixed. Then the read picture data in each small are is DCT-transformed through a DCT circuit 5 to obtain a compressed picture removed of high frequency components. Compressed picture data in each small area obtained like this is inputted to a neural network 6 along a horizontal scanning line. Then the category of plural edges and corners is decided.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technology for recognizing the shape of a very small component such as a chip mounter.

[0002]

2. Description of the Related Art Conventionally, as a method of recognizing the shape of a small component such as a chip by a chip mounter when mounting the small component, a captured image of an electronic component is converted into a grayscale image or a binary image. Convert to image and save,
The edge or corner image extracted from the saved image is subjected to run-length coding or chain coding to measure the characteristics of the edge or corner and determine the shape of the part by comparison. is there.

In such a conventional method, there is a problem that a large amount of information is processed by various operations, and a lot of time is spent for recognition of the component shape.

[0004]

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned problems of the prior art, and has been made in consideration of the above problems. It is an object of the present invention to provide a position recognition method.

[0005]

SUMMARY OF THE INVENTION The present invention is a method for encoding image data of a captured part image, inputting the encoded data to a neural network, and outputting a category from the neural network.

The categories are straight edge, convex edge,
These are concave edges, NG edges or straight corners, R chamfer corners, C chamfer corners, and NG corners.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a component shape recognition method according to the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block circuit diagram showing one embodiment of the apparatus. In the figure, reference numeral 1 denotes a CPU (Central
Processing Unit) bus and the CP
CPU 2, main memory 3, frame memory 4, DCT (Discrete Cosine Tra)
nsform) circuit 5 and a neural network 6 are connected.

[0009] Here, DCT is one of the image coding methods, and the image coding technology is a digital image communication method.
In the field of recording, it is indispensable for improving the efficiency of communication lines and recording media.

In particular, DCT extracts a sequence component (corresponding to a frequency component of Fourier transform) when a target data sequence is decomposed into a cosine waveform (for example, Image Lab., October 1990, pp. 41-45). Page, Journal of the Institute of Television Engineers of Japan, Vol. 43, No. 10 (1989) pp.
1455-1152).

Next, the component image obtained by the image input system 7 is temporarily stored in the frame memory 4, read out from the frame memory 4 at the time of DCT processing by the DCT circuit 5, and input to the DCT circuit 5. .

In the DCT circuit 5, as shown in the left diagram of FIG. 2 (a), all high-frequency components in the original image of the captured component are replaced with 0, and compressed data as shown in the right diagram is created.

Since an electronic component image contains many low-frequency components, even if a process for ignoring the high-frequency components after conversion is performed, the characteristics of the image are not largely lost. Therefore, even with a small number of data, sufficiently accurate data collation can be performed.

In the actual part image, FIG.
2B, the output of the DCT circuit is a power distribution image as shown in the right diagram of FIG. 2B. Such a power distribution image is obtained from the 0th sequence to the Mth sequence in the DCT expression represented by Expression 1.
-1 sequence (the higher the M, the higher the frequency)
It is arranged in a matrix in the Y direction,
Each sequence has a waveform as shown in FIG.

[0015]

(Equation 1)

The compressed data processed by the DCT circuit 5 is input to an input layer of the neural network 6, and a category of each edge or each corner is prepared in an output layer of the neural network 6, The edge or corner attribute of the captured part is specified by the magnitude of the output value for the category.

Next, the operation of the component shape recognition apparatus of the present invention will be described with reference to the flowchart of FIG. First, step S
In 1, an image of a component is input by the image input system 7.
The input image (512 × 512 pixels) is once taken into the frame memory 4, read out at an appropriate timing according to a command from the CPU 2, and an image of the edge or corner portion specified in step S 11 is extracted.

The data is output for each predetermined small area (64 × 64 pixels) for these edge or corner partial images (step S2). And step S3
The DCT circuit 5 performs DCT conversion on the image data of each small area read by the DCT circuit 5 to obtain a compressed image (32 × 32 pixels) from which high-frequency components have been removed.

The compressed image data for each small area thus obtained is input to the neural network 6 along a horizontal scanning line (step S4). An arbitrary row of compressed data of the horizontal scanning line is input as an input to the neural network 6, and the neural network is activated and the input scanning line data is converted into a straight edge, a convex edge, a concave edge, and an NG edge (these The degree of matching to match the edge that does not correspond to any of them is calculated, and which edge is determined as the one having the highest degree of matching (step S5).

The above-described scanning is sequentially performed for each horizontal scanning line. After processing all the component image data, a plurality of edge categories are determined. Similarly, a plurality of corner categories are determined.

When the component image is divided into small areas and DCT-transformed, the image data near the boundaries of each area becomes discrete, resulting in poor accuracy. For this reason, the DCT is performed by dividing each of the small areas from the original image so as to overlap each other and smoothing data near the boundaries of the respective areas.
It is desirable to give continuity to the compressed data of each area after conversion.

[0022]

As described above, according to the present invention, the neural network is activated by the data compressed by the DCT, and the edge or corner category is determined by the processing of the neural network. It is possible to save the memory by omitting it, obtain high accuracy of edge or corner shape recognition, and expect an effect that high-speed and high-precision component position recognition can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a hardware configuration of a device of the present invention.

FIGS. 2A and 2B are diagrams showing the concept of DCT processing.

FIG. 3 is a diagram showing a waveform (extract) of each sequence of the DCT.

FIG. 4 is a conceptual diagram showing a flowchart and status explaining the method of the present invention.

[Description of Signs] 1 CPU bus 2 CPU 3 Main memory 4 Frame memory 5 DCT circuit 6 Neural network 7 Image input system

Claims (3)

[Claims] 1. A component shape recognition method for encoding image data of a captured component image, inputting the encoded data to a neural network, and outputting a category from the neural network. 2. The category includes a straight edge, a convex edge,
The component shape recognition method according to claim 1, wherein the component shape is a concave edge or an NG edge. 3. The category according to claim 1, wherein the categories are a right angle corner, an R chamfer corner, a C chamfer corner, and an NG corner.
Or the component shape recognition method according to 2.